InsiderLLM
Archives
Search...
Log in
Subscribe
The chip that draws more power uses less of it
October 19, 2026
InsiderLLM Weekly issue 26 -- October 19, 2026 The ESP32-P4 board draws more power than the ESP32-S3, and it still spends about 46% less energy on every...
An $8 chip, twice as fast, same story
October 12, 2026
InsiderLLM Weekly issue 25 -- October 12, 2026 An $8 microcontroller now writes its children's stories more than twice as fast as it did out of the box, and...
We published someone else's benchmark. Here's the fix.
October 5, 2026
InsiderLLM Weekly issue 24 -- October 5, 2026 No new article this week. I spent it checking our own numbers. One benchmark on this site turned out to be...
A 27B for your 12 GB card. It lost one field.
September 28, 2026
InsiderLLM Weekly issue 23 -- September 28, 2026 A 27B now fits a 12 GB card and decodes 1.8x faster than the Q4 it was squeezed from. It also lost one field...
Stop writing the answer. Pick it. 47 items waiting.
September 21, 2026
InsiderLLM Weekly issue 22 -- September 21, 2026 No new article this week — the bench rig went up instead, so this issue is what is queued and what the logs...
Nineteen gigabytes of my 3090 sat empty. The 3060 kept up.
September 14, 2026
InsiderLLM Weekly issue 21 -- September 14, 2026 Last week I said Flash-Next was on the 3090 and I would not print a number until the 3060 box had its turn....
Nvidia's router dealt the cards evenly. That was the whole problem.
September 8, 2026
InsiderLLM Weekly issue 20 -- September 8, 2026 A day late: Monday went to the two bench runs at the bottom of this issue. Nvidia shipped a router on...
I got the result I wanted. Then I paid $4.97 to run it nine more times.
September 1, 2026
InsiderLLM Weekly issue 19 -- September 1, 2026 Twenty-four cents of API credit bought me a 91-line skill file, and it made the 27B on my desk 10.6 points...
Qwen 3.8 isn't slow. It's just very, very thorough.
August 24, 2026
InsiderLLM Weekly issue 18 -- August 24, 2026 14,953 tokens for one line of Python. Nine tokens for the identical line, from the identical file, with a...
OpenAI deleted the agents' message board. It didn't help.
August 10, 2026
InsiderLLM Weekly issue 17 -- August 10, 2026 OpenAI spent two days closing the channel its agents had been using to talk to each other. The agents spent two...
The offload cliff isn't a cliff. It's a straight line.
August 3, 2026
InsiderLLM Weekly issue 16 -- August 3, 2026 Every layer you push off the card costs roughly half a millisecond, and it keeps costing that all the way down —...
The whole Qwen 35B fits on a 3090. No offload at all.
July 28, 2026
InsiderLLM Weekly issue 15 -- July 28, 2026 Three things this week and they're the same thing: a number everybody repeats, us included, that nobody actually...
Hugging Face got hacked -- Open local AI came to the rescue!
July 20, 2026
InsiderLLM Weekly issue 14 -- July 20, 2026 Two stories this week look like opposites — one about models you can't run, one about a company getting broken...
Open weights are a weapon now. Yours still work.
July 14, 2026
The models on your drive didn't change this week — what they mean did. Two stories landed that look unrelated and aren't: China is funding open weights like...
The 'expensive' GPU came out cheaper — we rented both to find out
July 6, 2026
One firsthand cost guide I've wanted to write for a while, a floor report from the AI Engineer World's Fair, and a quick status check on the Qwen 3.7 open...
Qwen 3.7's open weights are overdue — by the math, not vibes
June 22, 2026
Two things landed this week, and one of them is a non-event that's actually the story: Qwen 3.7's open weights still haven't shipped, and by Qwen's own...
DeepSeek V4 Gets Real, and a Deprecation Trap to Dodge
June 15, 2026
InsiderLLM Weekly issue 13 — June 15, 2026 Mostly a DeepSeek week, and not for the reason you'd expect — the news isn't a new model, it's the unglamorous...
Ollama quietly changed how your Mac runs models
June 8, 2026
InsiderLLM Weekly issue 12 -- June 8, 2026 A quiet but real change in how Ollama runs on Apple Silicon, the model picks worth updating in your setup right...
A million-token model just beat Opus. You can't run it.
June 1, 2026
InsiderLLM Weekly issue 11 -- June 1, 2026 Big model launch this week with an asterisk, a real shift in what "local AI hardware" is about to mean on the...
Backend wars, Mac math, and the back-catalog refresh
May 26, 2026
InsiderLLM Weekly issue 10 -- May 25, 2026 The Qwen 3.6 ecosystem stopped being new this week and started being mapped. Three backends benched head to head...
Older archives